Transcript Slide 1

Basic research and economic growth
Working paper available on line at
http://hdl.handle.net/1887/18636
Cornelis van Bochove
Overview
• Growth Literature
• The technology function: applied R&D and
technology
• The hypothesis function: basic research.
• Return on applied R and D and basic research
• Neoclassical Science policy
Current growth literature
(Part of advanced macro economics)
Growth Literature: production function
Growth of A: “Technological progress”
The neoclassical theory of economic growth
• Robert Solow (1960): production function plus
saving-based capital accumulation  sustained per
capita growth
• Technological progress also needed (earned him the
1987 Nobel in economics; foundation of attention
for innovation)
• Exogenous (exponential) until 1990 (“Manna from
heaven”).
• From 1990 endogenous growth theory: tries to
explain technological progress
5
Endogenous growth: basic model
• Technological progress proportional to amount of R & D
– There is a stock of knowledge (“Ideas”) that grows
proportionally to amount of R&D.
– Technological progress = growth of stock of knowledge
• Then: constant level of R&D = constant rate of economic
growth.
• Higher level of R&D yields higher rate of economic growth.
• Stock of knowledge in 2008 version of UN-EU System of
National Accounts, R&D now officially investment
• Nice for science ministers and research councils
Endogenous growth: the scale problem
• But there is a problem: the scale problem
• Bigger economy spends more on R&D. Thus higher
growth rate.
• Growing population or higher R&D productivity
imply growing economy, hence more R&D spending
and higher growth rate.
• Thus more technological progress, still more spent
on R&D, ……. , economy explodes
Billionaires galore
Solution of scale problem: dimishing returns
• Knowledge accumulation progressively more
difficult
• Then growth of stock of knowledge no longer
proportional to amount of R&D
• Explosion vanishes, but growth too!
• Only some very slow growth if population grows
Solution of scale problem: need for knowledge
scale dependent
• Most used mechanism: product diversity increases
with scale, equal knowledge needed for all products
• Growing number of products plus fixed amounts of
R&D per product total R&D grows
• Relation between level of R and D and rate of
growth is back
•
But, but, but
Problems of solution
• Extreme fine-tuning: product diversity must grow at exactly
the same rate as the economy, or else …; and very sensitive
for unknown parameter values
• Disaggregation of R&D extremely difficult. Recent effort at
ETH to distinguish basic research complicated/artificial.
• Strict relation between country size and product diversity?
• Empirically: measurement tough and inconclusive
(Donselaar´s recent dissertation)
• The model is like Ashley and Pudsey
• Won Britain’s got talent 2012, not because they danced most
beautifully …
But because Pudsey can dance at all
Finance ministers know these problems. They don’t
like dancing dogs
12
Stock of knowledge reasonable assumption?
• Knowledge not cumulative, new results replace/
transcend old ones
• No measurable/definable stock of knowledge like
housing stock, capital stock, road network.
• Knowledge no factor of production in traditional
sense, stock-flow model does not work for
knowledge/ technological progress
• Can we generate endogenous growth without a
stock-flow model of knowledge?
13
Alternative, Part 1:
Technology and applied R&D
Trial and error
• Technological progress is trial and error process
• Solutions, ideas, possibilities tried out all the time.
• Those that work best: retained/put into practice
• Other ones/existing solutions: discarded/forgotten
• Formally: accumulation replaced by repeated sampling and
selection from a random distribution
• Analogous to Kortum, Econometrica, 1997
• I call this distribution the ‘Technology function’
The technology function
• Many technologies to produce the same basket of
goods and services
• They differ in efficiency/productivity
• Number of technologies smaller at higher
productivity level: more ways to work
inefficiently than efficiently.
• Thus technological possibilities have random
distribution that is skewed, with a long tail
• Like Kortum, I use inverse power law (= Pareto
distribution). Lots of those in scientometrics.
The Pareto technologyfunction
• Shape everywhere the same
• k is the power, low k, innovation easy
Pure trial and error (pre-industrial)
• Applied R&D = sampling (search) from the distribution
(“experimentation”), retaining the best result
• Technological process then depends on sampling process
• Simplest process: a-select trial and error until something
better than current practice is found
• Basis of pre-industrial innovation
• The higher the level achieved, the more additional
experimentation needed for further progress
• Thus slow growth, and only if means for experimentation
grow by population growth.
• Accords with available very long term growth data.
•
Corresponds to diminishing returns case, but now
with clear historical interpretation
Experimental learning
• Next suppose that by experimentation you not just find
better technologies, but also learn to experiment with
greater chance of success.
• Technology function the same, but lower boundary shifts
upward as a consequence of the experiments
China and other transitionals
• Learning by experimentation requires sufficient basic scientific
knowledge.
• Situation of transitional countries: China, India, Brasil, Japan
until 1980.
• There applications of experiments (including application of
licensed and copied technologies) generate more means for
further experiments, and
• render those successful too, as lower boundary shifts upwards.
• Thus double digit, explosive growth: Dagobert Duck
• Until they catch up and run into lack of basic knowledge. Cf.
Japan from 1980 on.
• Remember Newton the alchemist: spent half his life on
experiments we now know are pointless
Alternative, part 2:
basic research
Basic research: the hypothesis function
• Basic research is: developing hypotheses, testing whether they explain
experimental results, retain the best and continue. A la David Deutsch
• At every level of technology hypothesizing is needed until basic
knowledge explains results of experiments and further experiments with
lesser results can be avoided. No more looking for gold in test tubes.
• I define a hypothesis function as the distribution of basic research
hypotheses
• Trial and error hypotheses development/testing until the explanatory
power exceeds the current minimum of the technology function.
• Then the minimum increases and the game starts again at a higher level.
• For simplicity let shape of the hypothesis function also be Pareto with
the same power k as the technology function
• How fast can hypotheses be developed and tested? Assume:
• proportionality with amount of basic R and D labor/spending;
• fixed rate of learning of standard research worker. In every field
the first hypothesis requires the same amount of work as in
earlier fields, the next one a bit less, and so on.
Modern economic growth
• This yields a nice and globally stable rate of long
term growth.
• Value depends on two opposing forces: technology
power k and rate of learning in basic research.
• If k is high (innovation is difficult), growth is lower.
• If the rate of learning is high, growth is higher.
• Growth boosted a bit by labor growth, but even if
that is zero, per capita income still grows, due to the
learning in basic research.
• Conclusion: ‘scale free’ neoclassical endogenous
growth without funny assumptions
Return on applied R and D and basic research
24
Return on research spending (1)
• Growth rate does not depend on the proportions of income that are spent
on physical capital, applied R and D and basic research.
• Similar to a long standing result (Solow, 1956) in growth theory: rate of
saving does not influence the growth rate.
• But it does influence the level of income and consumption.
• And this is the same for research.
• Surprisingly easy to derive the rates of return once you have the model.
• Caveat: our model is for the world or the OECD as a whole
• They depend on the two parameters (technology power, rate of learning
in basic research) that we do not yet know.
• But they are not very sensitive to these unknown parameters.
• Rather, they are very sensitive to the current levels of the rate saving, the
applied R& D intensity and the basic R and D intensity.
Return on research spending (2)
• And these we know!
• For the rate of saving we may use 15 %, for applied R and
1.5 % and for basic research at most 0.5%. Then:
• The return on applied R&D spending is about ten times that
on physical capital investment and that on basic R &D about
three times higher still
• One euro extra applied R&D generates about 15 euro
national income.
• One euro extra basic research generates about 50 euro extra
national income.
• Hence: if the OECD governments spend one extra euro on
basic research, they will receive about 25 euro extra taxes
Neoclassical Science policy
Neoclassical Science policy: free or targeted
basic research (1)
• Long term economic growth of OECD countries
depends almost exclusively on basic research.
• Therefore basic research policy is very important
• First big question: should governments leave basic
research alone or try to direct it to specific sectors?
• This can be modeled by the allocation of research
labor to the basic research fields
28
Neoclassical Science Policy: free or targeted
basic research (2)
• Free research: all basic research labor works for all
technologies at the same time
• Targeted research: compartmentalization so that they
work mostly for a select number of technology fields
• I have modeled the two extremes: completely free
allocation and complete targeting (‘Topsectoren
beleid’)
• The latter halves the rate of economic growth
29
Neoclassical Science policy: effectiveness of basic
research (1)
• Rate of learning in basic research fundamentally
determines economic growth
• The main aim of science policy should be to raise
this rate of learning. This means:
• Open access to publications and data
• Excellent high speed research networks
• Ample research facilities (small and big), and easy
access to them more important than more
researchers!
• Excellent training for young researchers (PhD’s)
Neoclassical Science policy: effectiveness of basic
research (1)
• But also: reduction of learning losses due to
massive outflow of PhD’s and Post Docs: early
tenure track selection.
• Stimulation of independence of young talent (they
learn faster and are quicker to choose the new
approaches)
• Defragmentation within universities, much easier
multidisciplinary collaboration; reduction of
within-university invoice culture
• Probably many more things, let’s think about it.
Thank you very much!